Application of the Correct
Information Unit Analysis to the
Naturally Occurring Conversation
of a Person With Aphasia
Mary L. Oelschlaeger*
John C. Thorne**
University of New Mexico
Albuquerque

The Correct Information Unit (CIU) analysis for measuring the communicative
informativeness and efficiency of connected speech (Nicholas & Brookshire,
1993) was applied to the naturally occurring conversation of a person with
moderate aphasia. Results indicated that, in this instance, reliable CIU measures
could not be obtained. Intrarater reliability for CIU and %CIU was low, reaching
only 72%, and interrater reliability was never greater than 63%. However,
reliability of word counts was good. Post hoc analysis of rater disagreements in
application of the CIU analysis revealed that the majority (72%) resulted from
insufficiencies in the scoring rules that were originally designed to measure single
speaker connected discourse. Two descriptive categories of disagreements were
identified: interpretations of informativeness and absence of rules. The remaining
28% of disagreements were attributable to human error in the application of
scoring rules. Comparison of findings with previous research and implications for
future research are discussed.
KEY WORDS: aphasia, conversation, measurement, correct unit analysis

A

n ongoing challenge in clinical aphasiology has been the description of the language ability of individuals with aphasia. This
challenge has been approached primarily through the use of standardized testing instruments (Goodglass & Kaplan, 1972; Kertesz, 1982;
Porch, 1981, for example). However, the observation that everyday language ability may exceed that predicted by formal test scores has led
researchers to develop other tools that better capture the communicative success of persons with aphasia (e.g., Blomert, Koster, Van Mier, &
Kean, 1987; Holland, 1980; Lomas et al., 1989).

One recent approach toward this end is reported by Nicholas and
Brookshire (1993). In their study, these authors describe the Correct
Information Unit (CIU) analysis, a rule-based scoring system for
measuring two aspects thought to be important to communicative success: the informativeness (the degree to which the speech of the individual imparts the intended message) and efficiency (the rate at which
the message is produced) of language use. Using specific elicitation
stimuli (6 picture description, 4 requests for personal and procedural
*Currently affiliated with Northern Arizona University, Flagstaff
**Currently affiliated with Albuquerque Public Schools, Albuquerque, NM

636

Journal
of Speech,
Language,
and
Hearing
Research
Vol.
636–648• •June
June
1999 • ©American Speech-Language-Hearing Association
Journal
of Speech,
Language,
and
Hearing
Research
• •
Vol.
4242• •636–648
1999
1092-4388/99/4203-0636

Downloaded From: http://jslhr.pubs.asha.org/pdfaccess.ashx?url=/data/journals/jslhr/929262/ by a Universite Laval User on 06/07/2017
Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

information), a “connected speech” sample is obtained.
The CIU analysis of connected speech involves measuring the rate at which the speaker produced speech and
combining it with a derived measure: the %CIU. The
%CIU measure combines the total word count in the connected speech sample with the number of words in that
count which meet the specific criteria necessary to be called
a correct information unit. A correct information unit is
defined as “words that are intelligible in context, accurate in relation to the picture(s) or topic, and relevant to
and informative about the content of the picture(s) or
the topic” (Nicholas & Brookshire, 1993, p. 348).

segment of a network news program and were asked to
discuss its contents. Both sampling procedures were 7
minutes in length. The %CIU of conversational samples
was then compared with that of connected speech. Application of the CIU analysis to the conversation
samples in this study was reported as reliable, with
interrater reliability reaching a mean level of .88. Results indicated that although persons with aphasia
spoke with a higher %CIU rate in conversational
samples than in connected speech, their performance
in the conversation samples could be predicted from
their connected speech performance.

As part of their description of the CIU analysis,
Nicholas and Brookshire offer psychometric information
related to the reliability and stability of scoring. Specifically, intrarater and interrater reliability exceeded
90% (Nicholas & Brookshire, 1993), and scores were
noted to be reasonably stable over time (Brookshire &
Nicholas, 1994). These results led the authors to the
conclusion that the CIU analysis “may yield stable
baseline performance against which changes in connected speech with treatment or manipulation of experimental variables can be measured” (Nicholas &
Brookshire, 1993, p. 344). The authors go on to point
out, however, that the stability of an individual’s performance should not be assumed and that the application
of the CIU as a measure of change with treatment effects would require determination of stability over repeated assessments for any individual.

These findings seem to bolster the applicability of
the CIU analysis as a means of measuring the communicative informativeness and efficiency of everyday language use. However, Doyle et al. suggest that additional
research is needed to support this conclusion. By experimental design, the conversation samples in their
study were obtained with structured elicitation tasks.
Conversation was elicited as subjects were told when,
where, how long, and what to talk about with greater
(topic constrained) or lesser (topic open) specificity.
Thus, the experimental constraints used to obtain conversation samples in this study preclude the generalizability of their findings to other conversational sampling conditions.

Since this initial report, a number of research studies (Brookshire & Nicholas, 1994, 1995; Doyle, Goda, &
Spencer, 1995; Doyle, Tsironas, Goda, & Kalinyak, 1996;
Nicholas & Brookshire,1995) using the CIU analysis
have been conducted. Most relevant to this discussion
is one reported by Doyle, Goda, and Spencer (1995).
These authors attempted to expand the application of
CIU analysis to discourse types other than those in
Nicholas and Brookshire’s (1993) original study. They
added conversational elicitation tasks to the original 10
stimuli, noting that “measuring communicative informativeness and efficiency under conversational discourse conditions is perhaps the most ecologically valid
means of determining the interpersonal verbal communication abilities of adults with aphasia” (p. 130).
Doyle et al.’s study compared the performance of
20 persons with aphasia on Nicholas and Brookshire’s
discourse tasks with their performance on two conversational discourse tasks. Conversational samples were
collected in “simulated natural environments” (Doyle
et al., 1995, p. 131) and were of two different types.
The first sample was identified as “topic-open.” Subjects and their conversational partners were instructed
to “discuss anything they chose.” The second sample
was “topic-constrained”; subjects watched a 4.5-minute

A question not addressed in research on the CIU
analysis is whether it may be used to measure the communicative informativeness and efficiency of naturally
occurring conversation. As Doyle et al. note (1995), a
greater understanding of its applicability to authentic
communicative interactions is needed if the %CIU is to
be established as an ecologically valid tool for measuring
communicative informativeness and efficiency of everyday language ability of persons with aphasia.
This study was designed as an initial effort toward
this end. The purpose was to answer two basic questions: (1) Can the CIU analysis described by Nicholas
and Brookshire (1993) for measuring communicative
informativeness and efficiency of connected speech be
applied reliably to the naturally occurring conversation
of a person with aphasia? and (2) If it can be reliably
applied, is the communicative informativeness and efficiency it measures a stable feature of the conversation
of a person with aphasia across time and across conversational settings?

Methodological Issues
In the design of this study, several methodological
issues were considered that related to research design
and determination of reliability and stability of the CIU
measure. The rationale for methodological decisions relating to these issues follows.
Oelschlaeger & Thorne: Application of the Correct Information Unit

Downloaded From: http://jslhr.pubs.asha.org/pdfaccess.ashx?url=/data/journals/jslhr/929262/ by a Universite Laval User on 06/07/2017
Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

637

Research Design
In both of their articles examining the stability of
their measures of informativeness and efficiency
(Brookshire & Nicholas, 1994; Nicholas & Brookshire,
1993) and in Doyle et al.’s (1995) study, group designs
were used. As Brookshire and Nicholas (1993) note, “Results based on group averages…do not speak directly of
the test-retest stability of individual subjects’ scores” (p.
402). They point out that clinical application of the CIU
analysis requires individual assessment of reliability and
stability of measures. Because of the importance of individual performance in clinical application, the conversation of a single subject was used for CIU analysis in
this study. The limitation of generalizability of findings
associated with study of a single case was accepted as
compatible with the design of this study as an initial
inquiry of the application of the CIU analysis to naturally occurring conversation.

Reliability
In all the studies utilizing the %CIU to date, each
has had a primary rater score the entire discourse sample
and then rescore a percentage of that sample to establish
intrarater reliability. Interrater reliability was established
in these studies by having two speech-language pathologists score a percentage of the discourse sample for comparison with the scores by the primary rater. This precedent was followed in the current study. Nicholas and
Brookshire (1993) also report that their procedure included a chance for the raters to sit down and compare
notes on how they were approaching their task after
having scored a percentage of the connected speech
sample. Doyle et al. (1995) do not report inclusion of
such practice sessions in their study of structured conversational sampling. As the purpose of this study was
to determine the reliability of the rule-based system for
discourse for which it was not designed, any discussion
that would potentially clear up ambiguity arising from
differing interpretations of those rules could introduce
a confounding element into this study. A test of reliability under these conditions would not be a test of the
rules as published. Rather, it would be a test of the rules
as published when augmented by the unwritten rules
agreed upon through a discussion between raters. On
the other hand, not including such opportunity could
potentially detract from the clinical meaningfulness of
this study, as it is common for clinicians to familiarize
and even practice assessment tools before administration. Because of these contrasting concerns, the decision was made not to include any formal opportunity
for practice or discussion but to provide raters with instructions to familiarize themselves with the rules and
to take as much time as needed to complete the task.

638

This provision allowed for raters to independently practice rule application. More specifically, no time constraints were placed on the raters’ completion of their
scoring of conversations (see Instructions 7 and 8 in the
Appendix). The elapsed time for completion of scoring
of all three raters was monitored and extended for as
long as 9 hours. The length of time taken to complete
the task as well as the extent and complexity of rules
(see Appendix B, Nicholas & Brookshire, 1993, pp. 348350) serve as indirect indices of the extent to which raters incorporated familiarization and independent practice as part of their participation in this study.
An additional issue related to the question of reliability is determination of the degree of rigor that is
required before a measure is said to be reliable. There
is currently no universally accepted standard for making this decision. (An extensive discussion of reliability
standards is beyond the scope of this article. The interested reader is referred to Pedhazur & Schmelkin, 1991
and/or Seibel, 1968). There are guidelines in the literature, however, that suggest that relatively low
reliabilities are acceptable for “ground breaking” research, that higher reliabilities are needed to determine
differences between groups, and that only very high
reliabilities [can] be used to make decisions about individuals (Pedhazur & Schmelkin, 1991).
The CIU analysis being examined in the current
study is proposed to be both a research and a clinical
measure. Important to clinical application specifically
is that it is designed to be used in making clinical decisions regarding patient progress, continued treatment,
and/or treatment protocols. Because clinical decisions
like these affect individuals’ lives, clinical measures must
meet high reliability standards. The reliability of 90%
found for the %CIU in Nicholas and Brookshire’s (1993)
original study and Doyle et al.’s report of 88% certainly
are adequate for clinical decision making. However, because the current study applies the rule-based system
to naturally occurring language phenomena for which
it was not designed, it would be unrealistic to demand
equally high reliability. On the other hand, a recent study
by Crockford and Lesser (1994) of other measures of the
everyday communication ability of persons with aphasia suggests that reliability in the low seventies is too
low for clinical application. These authors examined the
use of several clinical tools to validate efficacy of treatment. Included in their examination was the Communicative Effectiveness Index (CETI) (Lomas et al., 1989).
The conduct of their study involved the administration
of CETI by more than one scorer. Results indicated that
the CETI had the least clinical utility for measuring
change when compared to scores derived from other
measures. The authors note that Lomas et al. report an
interrater reliability of .73 (with a confidence interval
of 95%) and suggest that this reliability level may be

Journal of Speech, Language, and Hearing Research • Vol. 42 • 636–648 • June 1999

Downloaded From: http://jslhr.pubs.asha.org/pdfaccess.ashx?url=/data/journals/jslhr/929262/ by a Universite Laval User on 06/07/2017
Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

insufficient to support the clinical application of this tool.
Given the findings of Crockford and Lesser, which suggest that 73% is too low and the expectation that the
90% reliability obtained in other CIU studies may be
too high, reliability scores greater than 80% were considered acceptable for the purposes of this study.

Method
Subject
Ed, a 50-year-old right-handed male, 6 years post
stroke with history of a single left hemisphere CVA with
residual moderate aphasia and mild right hemiparesis,
served as the subject for this study. He was selected from
among members of the local stroke club who volunteered
for participation in research because of his similarity with
the subject selection criteria of Nicholas and Brookshire
(1993). Specifically, his Porch Index of Communicative
Ability (PICA; Porch, 1981) percentile score was 62, which
conformed to the aphasia severity of Nicholas and
Brookshire’s “average” subject (SPICA %ile average
equaled 63.7). His Western Aphasia Battery (WAB;
Kertesz, 1982) Aphasia Quotient was 46.6, with a WAB
classification of conduction aphasia. His speech was intelligible, and he had sufficient language ability to actively
participate in conversation. Ed was able to understand
what was being said to him relatively well and was generally able to follow the topic of conversation. However, at
times when comprehension was dependent on a single
word or when the message was too long or linguistically
complex, he would misunderstand. Although he was able
to use many different forms of language and could almost
always get his main idea across, his spontaneous speech
was replete with instances of word-finding difficulty.
Conversational partners included (in all conversations and settings) his wife of 28 years, M, as well as (in
multiparty conversations) the first author, MO, and her
graduate research assistant, MG. All three conversational participants were ordinary speakers with no history of illness, disease, or deficit.

speakers (e.g., Goodwin, 1987; Goodwin & Goodwin, 1986),
persons with Alzheimer’s disease (Orange, Lubinski, &
Higginbotham, 1996), and persons with aphasia (Goodwin, 1995; Oelschlaeger, 1999; Oelschlaeger & Damico,
1998a, 1998b; Simmons-Mackie & Damico, 1996). Procedures included the videorecording of naturally occurring conversations over an extended period of time in a
familiar setting. An 8-mm video camera was left in the
subject’s home for approximately 6 weeks. The couple
was taught how to use the equipment and told to use it
to record their conversations at their own discretion. No
further direction was given as to when, where, how often, or in what manner of conversations were to be recorded. This resulted in the capturing of 5 conversations
between the subject and his wife, recorded outdoors on
their backyard patio and totaling 149 minutes of conversation. In addition to these dyadic conversations, 3
multiparty conversations were recorded in the couple’s
home. Conversational partners included—in addition to
Ed and M—MO (two sessions) and MG (in all three conversations). These conversations took place with participants seated around the couple’s dining room table
and were allowed to develop naturally with no preagreed
topic, length, or procedural arrangements. These conversations resulted in 117 minutes of taped discourse.
In total, 266 minutes (4 hours and 26 minutes) of
videorecorded conversation were obtained.
Videotaped conversations were viewed and transcribed by trained research assistants. Conversations
were then reviewed many times by the authors while
reading from these transcripts. Discrepancies between
the video and the initial transcripts were resolved by
consensus and corrections made. The transcript and videotapes were used to segment the conversations into
turns (referred to as Turns at Talk, TAT, in this report)
in keeping with the standard practice in the study of
naturally occurring conversation (see, for example,
Levinson, 1983; Orange et al., 1996; Sacks, Schegloff, &
Jefferson, 1974).

Data Analysis

Data Collection

Conversation Participation

Conversations used in this study were collected as
a part of a larger, on-going study of the conversational
strategies of individuals with aphasia and their spouses
(Oelschlaeger, 1999; Oelschlaeger & Damico, 1998a,
1998b).

Information about Ed’s participation in the eight
conversations is presented in Table 1. As noted in Table
1, Ed took a total number of 1333 conversational turns
and spoke a total of 8113 words in these conversations,
an average of 6 words per turn. Of his turns 62% (830)
were multiword utterances, an average of 104 per conversation. These findings indicate that he was an active participant in the conversations. More importantly,
the number of conversations and the magnitude and
the nature of his participation demonstrate how well
these conversations represent Ed’s naturally occurring

To ensure authenticity, conversation collection procedures were conducted according to the commonly accepted protocol for the study of naturally occurring conversation (e.g., Atkinson & Heritage, 1994; Levinson,
1983). These procedures are well established in research
on the naturally occurring conversation of ordinary

Oelschlaeger & Thorne: Application of the Correct Information Unit

Downloaded From: http://jslhr.pubs.asha.org/pdfaccess.ashx?url=/data/journals/jslhr/929262/ by a Universite Laval User on 06/07/2017
Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

639

Table 1. Parameters of Ed’s participation in conversation.

Conversation
A*
B
C*
D
E*
F
G
H

42 min
28 min
31 min
35 min
44 min
19 min
31 min
36 min

Total: 266 min

Turns at
Talk

Words per
conversation

Average # words
Turn at Talk**

Multiword
Turns at Talk

Time
talking***

211
212
169
186
184
64
122
185

1595
1208
1041
749
1558
272
438
1252

8
6
6
4
9
4
4
7

124
150
104
112
105
35
65
135

17 min
12 min
13 min
8 min
15 min
4 min
4 min
16 min

Total : 1333

Total: 8113

Average: 6

Total: 830

Total : 89 min

*Multiparty conversations (All others were two-party conversations.)
**Rounded off to whole number
***Rounded to nearest minute

everyday language and, as such, offer objective support
for authenticity of conversations.

Timing of Conversational
Participation
Ed’s Turns at Talk (TAT) in conversations were timed
to provide temporal data needed for the efficiency calculations of the CIU analysis. However, because conversations are dialogic and naturally contain extended
periods of time when no one is talking, the method used
by Nicholas and Brookshire (1993) required modification. It was apparent that simply subtracting all contributions for participants other than the subject from the
total time would grossly overestimate the amount of time
Ed took to say what he had to say. For this reason, a
method for timing Ed’s talk was used that eliminated
extended silences from the data set. Specifically, Ed’s TAT
were timed using a hand-held stopwatch. Pauses that
occurred within the boundaries of his TAT were included
in the total time for that TAT. Pauses that occurred outside TAT boundaries were not included. Short, single-word
TAT or extremely short, multiword TAT (i.e., “Oh yeah”)
were given a standard time score of .25 seconds, as this
was the lower time limit for which the watch being used
could be manually started and then stopped. Ten percent
of Ed’s TAT were retimed as an intraexaminer reliability
measure. Point-to-point reliability of TAT times was
90.19%, with a margin of error of ±.10 of a second.

Reliability of Word and CIU Counts
A complete data set consisting of word count, CIU
count, and the derived %CIU was obtained from the
application of the rule-based system for CIU analysis to
Ed’s talk in all conversations by the primary rater (the
second author). Intrarater reliability calculations were
640

generated 4 weeks after initial counts were made. At
this time, the primary rater rescored Ed’s talk in a randomly selected conversation from each conversational
setting (one dyadic, one multiparty). Ed’s TAT in these
two conversations represent approximately 25% of the
TAT available for analysis.
Data for interrater reliability were generated by
having two speech-language pathologists (Raters One
and Two) score Ed’s TAT in the two randomly selected
conversations used in the intrarater reliability calculations. Both speech-language pathologists serving as independent raters in this study were ASHA-certified and
had been professionally involved in the diagnosis and
treatment of persons with aphasia for over 3 years. Both
were currently employed, one in a rehabilitation setting
and the other as a university clinical supervisor of graduate students clinically involved with neurologically impaired adults. Neither Rater One or Two had any previous interaction with the subject.
Raters were told they were participating in a study
examining the applicability of a standardized measure of
conversation. They were given a packet containing a full
transcript (i.e., transcripts included the conversational
turns of every participant) of each conversation. They were
also given a copy of the rules for making word and CIU
counts and a set of written instructions specific to the procedures of the study (see Appendix). Scoring was done
independently by each rater in a controlled environment.
Both interrater and intrarater reliability for word
and CIU counts of Ed’s TAT were calculated by dividing
the number of agreements by the sum of the agreements
and disagreements and multiplying by 100. A point-topoint reliability calculation was done. These calculations
were made on a TAT-by-TAT basis, rather than a wordby-word basis, as word count was one of the measures
being studied. Specifically, the number of CIUs and/or

Journal of Speech, Language, and Hearing Research • Vol. 42 • 636–648 • June 1999

Downloaded From: http://jslhr.pubs.asha.org/pdfaccess.ashx?url=/data/journals/jslhr/929262/ by a Universite Laval User on 06/07/2017
Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

words for each turn at talk for each rater was totaled.
These totals were then compared for agreement across
raters. Agreement was considered to exist when equal
scores were given by raters for the total number of words
or CIUs in a particular TAT.

Table 2. Intrarater and interrater reliability for all measures by
rater.
Word
count (%)

Rater

CIU
count (%)

%CIU

72
71.96
72.63

73
72.43
72.63

Intrarater reliability

Results
Reliability of Word Count,
CIU Count, and %CIU
Intrarater and interrater reliability for all measures,
raters, and conversations is reported in Table 2. Data
are reported first as an average reliability score for the
entire data sample. Data on reliability found in each
conversational setting are then reported so that comparison across conversational settings can be made.

Word Count Reliability
As seen in Table 2, both the intrarater and interrater reliability of word count were found to be high
(greater than 90%) for all raters across both conversation settings.

CIU Count Reliability
Conversely, both intrarater and interrater reliability for CIU counts was poor (less than 73% for intrarater
and less than 56% for interrater). Conversational setting had no apparent impact on these scores, as scores
from each setting fell within five percentage points of
each other.

%CIU Reliability
The %CIU is a calculated measure combining word
count and CIU count. Its reliability is, therefore, affected by the reliability of the measures used in its calculation. Reliability for %CIU was calculated on an
agreement/disagreement basis. Intrarater reliability for
%CIU was higher (73%) than that for interrater reliability (never greater than 55%). Both fell short of an
acceptable range of 80%, the established criterion level
for the purposes of this study. Though the variation
between scores in the different conversational settings
was greater for %CIU than for the other measures, no
pattern emerged that would indicate a consistent influence of conversational setting on the reliability of
the measure.
During analysis of the reliability data, it was noted
that within individual TAT, different types of disagreement occurred between the different raters. For example,
two of the raters may have agreed on the number of
words in a particular TAT, but disagreed on the number
of CIUs. On this same TAT, the third rater may have

Author
Total sample
Dyadic conversation
Multiparty conversation

98
99.1
97.8

Interrater reliability
Author to Rater One
Total sample
Dyadic conversation
Multiparty conversation

98
96.26
99.12

56
57.28
54.19

55
57.01
53.07

Author to Rater Two
Total sample
Dyadic conversation
Multiparty conversation

91
91.59
90.05

56
53.27
58.66

55
52.8
57.54

Rater One to Rater Two
Total sample
Dyadic conversation
Multiparty conversation

91
90.19
91.06

52
52.34
50.84

51
51.4
50.28

Average across all raters
and both conversations

93

54

54

disagreed with the other raters’ word count, but agreed
with the CIU count of one of the raters. To get a more
accurate picture of the degree to which this was occurring, word and CIU counts were analyzed to examine
what percentage of TAT were judged to have equal
numbers of both word and CIU counts by all three raters. This analysis found a consensus among raters for
only 40.65% of TAT in the dyadic conversation and for
37.43% of TAT in the multiparty conversation. This resulted in a consensus for both word and CIU count for
only 39% of TAT in the overall reliability sample.

Stability of Measures
Assessment of the stability of measures was designed according to the recommendations of Brookshire
and Nicholas (1994). That is, Ed’s TAT were segmented
into 300 word samples to allow correlational comparisons (mean, standard deviation, variance, and Pearson
Product Moment) of both speech rate and %CIU to be
made across time and conversational settings. As reliability of the measures was not established, stability
calculations were not made. Nonetheless, it was possible to determine the stability of speech rate because
these calculations had been made in preparation for
determination of the stability of the %CIU measure.
Results are graphically presented in Figure 1. Speech
Oelschlaeger & Thorne: Application of the Correct Information Unit

Downloaded From: http://jslhr.pubs.asha.org/pdfaccess.ashx?url=/data/journals/jslhr/929262/ by a Universite Laval User on 06/07/2017
Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

641

Figure 1. Ed’s average speech rate per conversation.

rate was noted to be relatively stable, averaging 91 words
per minute with a range of 68 to 110 words per minute.

Post Hoc Analysis
The results reported above paint a rather complicated picture. Although it is clear that reliable scores
could not be obtained for both CIU counts and %CIU
when the rule-based system for CIU analysis was applied to the naturally occurring conversation of a person with aphasia, the reasons for the lack of reliability
are not apparent in the data. As Porch (personal communication, 1997) observes, there are two primary
causes for poor reliability: incompetence of the raters or
insufficiency of the scoring system. Given the desire in
clinical aphasiology to develop a way to capture the communicative informativeness and efficiency of everyday
language and the high reliability reported in all the other
published studies using the CIU analysis, additional
analyses were performed. Specifically, post hoc analyses
were conducted to explore whether the raters were not
competent raters or the rule system was not sufficient
in some way and, therefore, did not allow raters to make
reliable judgments when analyzing the naturally occurring conversation of an individual with aphasia.

They rescored a randomly selected portion of TAT equal
to approximately 20% of the original sample approximately 6 weeks after their initial scoring sessions.
Intrarater reliability was then calculated for each rater
using the formula and methods described above. Results
of this analysis are presented in Table 3. Data are reported first as an average reliability score for the entire
data sample. Data on reliability found in each conversational setting are then reported so that comparison
across conversational settings can be made.
As noted in Table 3, intrarater reliability of the independent raters for word count was good, averaging
86%. This is considered to be within a range acceptable
for most purposes. However, the intrarater reliability
scores for both CIU count and %CIU were not within an
acceptable range, reaching only 61% and 57% for Rater
One and 54% and 53% for Rater Two respectively. Again
no consistent pattern emerged regarding conversational
setting in relation to these scores. These findings, combined with those of the primary analysis (see Table 2),
show inconsistency in application of the CIU analysis
rules within and across all raters.
Accuracy of rule application. The second approach
Table 3. Intrarater reliability for independent raters.

Competence of the Raters
The question of the competence of the raters was
approached in two ways. The first involved analysis of
the intrarater reliability of the independent raters. The
second involved analysis of the accuracy of application
(e.g., human error) of the rules by all three raters.
Intrarater reliability of raters. To determine the consistency with which these independent raters applied
the rules for making word and CIU counts, the two
speech-language pathologists were asked to rescore a
portion of the conversations they had scored originally.
642

Rater

Word
count (%)

CIU
count (%)

%CIU

Rater One
Total sample
Dyadic conversation
Multiparty conversation

85
90.57
79.41

61
56.6
64.71

57
54.72
58.82

Rater Two
Total sample
Dyadic conversation
Multiparty conversation

87
88.68
85.29

54
58.49
50

53
56.6
50

Journal of Speech, Language, and Hearing Research • Vol. 42 • 636–648 • June 1999

Downloaded From: http://jslhr.pubs.asha.org/pdfaccess.ashx?url=/data/journals/jslhr/929262/ by a Universite Laval User on 06/07/2017
Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

to investigating rater competency involved examining
the accuracy with which all raters applied the rule-based
system for CIU analysis. The rules for scoring and counting words and correct information units (CIUs) are published in Appendix B in Nicholas and Brookshire’s 1993
article (pp. 348–350). The rules are extensive, text-based,
and complex as they include inclusion and exclusion criteria to be applied to scoring of speech as a correct information unit.
To determine accuracy of rule application, all TAT
for which there was disagreement between raters were
identified by the second author. In total, 350 discrepancies were noted. Then, the discrepancies were categorized as conforming to the rules or as a misapplication
of an unambiguous rule (i.e., human error). For example,
in the dyadic conversation, one rater determined that
one of Ed’s TAT, “And Norma’s is two,” had 4 CIUs, counting And at the beginning of the TAT as a CIU, whereas
the other raters counted only 3 CIUs, excluding And
from their count. As the word and is excluded from all
CIU counts according to Nicholas and Brookshire’s
(1993) Rule 2.20 (“And is never counted as a correct
information unit…,” p. 349), the disagreement between
raters on that TAT was attributed to human error. Evaluation of the reliability of this categorization involved the
independent categorization of the 350 discrepancies by
the first author. The proportion of agreement was 96%,
using the formula of agreements divided by agreements
plus disagreements times 100.
Results of this analysis were that 99 of the 350 discrepancies were attributable to misapplication of an unambiguous rule to the TAT by one or more of the raters.
These 99 instances accounted for just over one quarter
(28%) of the disagreements between raters. An easy majority of these errors fell into two categories: counting and
as a CIU and counting vague words such as thing or stuff
as CIUs. Another commonly occurring error involved one
rater’s misapplication of the rules that define whether or
not a conjunction is a CIU. (This rater excluded nearly all
coordinating conjunctions from her count, overgeneralizing a rule that excludes conjunctions used as filler.)
Clearly, human error influenced the reliability noted
in this study. To determine the extent of this influence,
these 99 disagreements were eliminated from the data
set, and interrater reliability was recalculated. The resulting reliability was 62.19, which again was below the
criterion-level agreement set in this study and indicated
that other factors were affecting reliability.

Sufficiency of the Rule-Based
System
To examine the sufficiency of the rule system for
application to naturally occurring conversation, an
analysis was performed on the remaining 251 (72%) TAT

where raters disagreed and where the disagreement
could not readily be attributed to a misapplication of a
rule. For example, in the multiparty conversation, one
of Ed’s TAT, “I would say, Italian was Italian. Go, there
were some Mexicans, but maybe not many,” was determined to have 15 CIUs by one rater who excluded only
the word Go in accordance with Rule 2.13, which excludes “dead end, false starts or revisions in which the
speaker begins an utterance but either revises it or
leaves it uncompleted and uninformative with regard
to the picture(s) or topic” (Nicholas & Brookshire, 1993,
p. 348). Another rater credited Ed with only 9 CIUs in
this same utterance. She excluded from her count “I
would say” and “Italian was” in addition to the word
Go. An examination of each of the disputed pieces of
this TAT revealed at least two rules that could be applied in order to determine if they were correct and
informative. For example, the phrase “I would say” could
easily be interpreted as excluded according to rule 2.21.
This rule excludes “commentary on the task and lead-in
phrases that do not give information about the picture(s)
or task and are not necessary for the grammatical completeness of the statement” (Nicholas & Brookshire,
1993, p. 349). If, however, a rater interpreted the phrase
“I would say” to be equivalent to the phrase “I would
estimate” or similarly “It is my opinion that,” then the
phrase could not be excluded by the rule. In this case
the phrase would be thought to convey relevant, correct
information to the listener and would, therefore, be considered as correct in context (Rule 3.11) and counted as
3 CIUs. Because both positions can be argued reasonably from the rules, the disagreement between raters
was attributed to rule insufficiency rather than human
error on the part of the raters, and the linguistic and
conversational context in which the disagreement occurred was noted. When this had been done for each of
the disagreements in both conversations, these various
linguistic and interactional features were classified in
order to examine patterns or trends in the data. This
led to the identification of two broad categories into
which these 251 discrepancies could be placed. The first
broad category related to the raters’ interpretations of
informativeness; the second related to the absence of
rules covering specific interactive language use found
in the conversations.
Interpretations of informativeness. Of these two categories, interpretations of informativeness was the largest, accounting for 159 (74%) of the 251 disagreements
that were not accounted for by human error. In this category, disagreements between raters seemed to stem
from the subjectivity involved in applying the central
principle of correctness used in the CIU analysis to the
complex interactive environment found in conversation.
The scoring rules indicate that words and phrases need
to be “correct in context,” even if not grammatically well
Oelschlaeger & Thorne: Application of the Correct Information Unit

Downloaded From: http://jslhr.pubs.asha.org/pdfaccess.ashx?url=/data/journals/jslhr/929262/ by a Universite Laval User on 06/07/2017
Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

643

formed, if they are to be counted as a correct information unit. This presupposes that a standard of correctness is discernable in the sample being scored. Nicholas
and Brookshire (1993) selected elicitation stimuli to constrain the types and topics of the connected speech they
sampled. However, in natural conversation, the types of
discourse present and topics covered are collaboratively
determined by participants (Sacks, 1992). For this reason, there is a much greater ambiguity from moment to
moment regarding exactly what direction the conversation will take. This inherent ambiguity affects the determination of correctness by a rater attempting to
analyze the conversation, particularly when one or more
of the conversational participants is a disordered
speaker. In the current study, disagreements resulting
from these factors were found either in TAT that contained a significant amount of revision and/or grammatical error or in TAT with a degree of ambiguity sufficient to support multiple interpretations. Because the
full transcript was provided, each rater had information available to them from the conversation context
(e.g., turns at talk that were antecedent and subsequent
to Ed’s turn) to use in making a determination of the
informativeness of Ed’s talk. However, because the rule
system was designed to measure single-speaker discourse, it does not provide guidelines on how this interactive information may be used. Therefore, each rater
approached the job of determining just which words in
Ed’s TAT were correct and informative differently. As a
result, each rater’s count of which words in a particular TAT conformed to the definition of a CIU was also
different.
An example from the data will help to clarify how
this occurred. In the dyadic conversation, Ed said, “Now
we, now we’re clean all the things that uh, vacuuming,
now we, all we have to do is this.” One rater excluded
almost this entire TAT from her CIU count, giving Ed
credit for “…we’re…vacuuming…” and scoring his TAT
at 3 CIUs. Another rater excluded much less of this TAT,
giving Ed credit for “…now we’re clean…vacuuming…all
we have to do is this” and scoring his TAT at 12 CIUs.
The third rater excluded still less of this statement giving Ed credit for “…now we’re clean all the…vacuuming,
now… all we have to do is this” and scoring his TAT at
15 CIUs. Although it is impossible to say exactly what
logic each rater used to decide which of the words in
this 20-word TAT provided correct information to listeners, it is clear that their interpretations of Ed’s
message were slightly different. The first rater seems
to have decided that Ed provided the information “we
were vacuuming.” The second rater interpreted Ed’s
message differently, deciding it contained more correct
information, something to the effect that “we’re clean,
have done the vacuuming, and all we have left to do is
this.” The third rater’s interpretation of the message
644

was similar to the second rater’s, but allowed for more
of it to be considered informative. In this case the
message could be construed to have been something
along the lines of “Now that we are clean, having done
all the vacuuming, all we have left to do is this.” When
each of the raters had decided which words conveyed
correct information in the context of each rater’s particular interpretation of the message, the rest of the TAT
was excluded from their CIU count. The determination
of what was “correct in context” in this instance was
quite different for each rater. As a result their CIU counts
were different.
Absence of rules. The second broad category of disagreement between the raters, accounting for 92 (just
over 26%) of the 251 disagreements between raters not
attributable to human error, resulted from absence of
rules for defining CIUs for the unique interactive language use that occurs in conversation and not in singlespeaker connected speech. Data analysis identified six
subcategorizes of TAT with specific linguistic and contextual features not addressed in the rule-based system
for CIU analysis. An example of each subcategory is presented in Table 4.
As displayed in Table 4, the subcategories relating
to absence of rules were (a) yeah and equivalent
affirmatives (such as right and uh huh) as an answer to
a direct question, (b) yeah and other interactive tools
(such as uh huh and hmm) used by Ed to acknowledge
or encourage the speaker, (c) automatic speech such as
social greetings or idiomatic phrases (what the hell and
maybe, maybe not, for example), (d) direct questions, (e)
repetition of another speaker’s remark to show agreement or acknowledgment, and (f) incomplete utterances
resulting from interruption.

Discussion
The purpose of this study was to address two primary questions: (1) Can the rule-based system for CIU
analysis designed by Nicholas and Brookshire (1993) for
measuring communicative informativeness and efficiency of connected speech be applied reliably to the
naturally occurring conversation of a person with aphasia? and (2) Is the communicative informativeness and
efficiency it measures a stable feature across time and
across conversational contexts? The answer to the first
of these questions was that no, as a whole, the CIU
analysis could not be applied reliably to the conversation of the single subject of this study. The answer to
the second question was precluded by the reliability findings, although speech rate was noted to be stable across
conversations.
Two competing possible explanations for the low
reliability noted in this study were identified. One was

Journal of Speech, Language, and Hearing Research • Vol. 42 • 636–648 • June 1999

Downloaded From: http://jslhr.pubs.asha.org/pdfaccess.ashx?url=/data/journals/jslhr/929262/ by a Universite Laval User on 06/07/2017
Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

Table 4. Subcategories and examples of absence of rules.
Example
Subcategory

Turn No.

Speaker

1. Yeah as an answer to a direct question.

274
275

MO
Ed

You mean for your person? Or for
//Yeah.

2. Yeah as an interactive tool.

63
64

M
Ed

It’s a pretty day.
Yeah.

3. Automatics.

75
76

M
Ed

The wall doesn’t look good, does it?
No, but what the hell.

4. Questions.

43
44

M
Ed

There it goes. Two-ton. Kitty, kitty, kitty.
Him?

5. Repetition of other speakers.

383
384
385

MG
M
Ed

Now that’s not all the bleachers.
That’s it.
That’s it.

6. Incomplete utterance due to
interruption by other speaker.

96

Ed

97

M

OK. Approval and everything. But the
card is the
//the American Express
or the credit card.

a

a

Utterance

MO = first author. M = wife. MG = research assistant.

that the raters were not up to the task. The other was
that there were insufficiencies in the rule-based system
which precluded its reliable application to naturally occurring conversation. The majority of evidence indicated
that the observed low reliability stemmed primarily from
rule insufficiency rather than rater competence. Direct
evidence relating to rater competence was noted in examination of the reliability of word counts. Specifically,
the high reliability for word counts (intrarater and
interrater scores of 93% and above) indicated that when
given a sufficiently clear set of rules, raters made clear
and consistent judgments regarding the fit of the data
to those rules. Less direct but supportive evidence of
rater competence was gained in analysis of errors resulting in discrepancies between raters. Specifically,
human error did occur but accounted for only 28% of
the total 350 discrepancies. Also, even if human errors
were totally eliminated from the data set, the overall
interrater reliability for CIU of 69.12 would still fall below the criterion of 80% set for the purposes of this study.
Finally, support for identifying raters as competent (i.e.,
capable of producing reliable scores) was gained from
the dissimilarity between CIU reliability and wordcount reliability. If the raters’ ability to make consistent
judgments were the only factor affecting reliability, then
it would be expected that reliability scores for the two
primary measures (word and CIU counts that are used
to calculate %CIU) would be at least similar. This was
most definitely not the case; CIU reliability scores fell
as much as 40 percentage points below Word Count reliability scores.

Even though the majority of evidence suggested
rater competence, human error did occur; and the issue
of raters’ practicing scoring rules before applying the
CIU analysis to any language sample warrants mention in this regard. The procedures for determining reliability in this study followed those used in Nicholas and
Brookshire’s (1993) initial report of the CIU analysis in
all aspects but one: No opportunity for discussion between raters relating to the rule system was provided.
However, an opportunity for “on the job training” of raters was provided. Although reliability for CIU was below acceptable levels for all raters, the primary rater
(second author) had higher intrarater reliability scores
than the other two raters. Given that the primary rater
had greater exposure to the CIU rules through his participation in the design and conduct of this study and
the scoring of all eight conversations, the possibility
arises that these experiences constituted a greater “selftraining” opportunity. In addition, the existence of 99
discrepancies between raters that were readily attributable to human error suggests that at least some improvement in reliability (albeit not enough to meet the
reliability criteria of this study) would be gained if raters practiced the CIU administration before applying it
to naturally occurring conversation. Although beyond
the scope of this study, the importance of practice merits consideration in future research.
The evidence that insufficiencies in the rule system
played a primary role in the poor reliability of the data
collected comes from the finding that a full 72% (251) of
the 350 disagreements between raters were attributable
Oelschlaeger & Thorne: Application of the Correct Information Unit

Downloaded From: http://jslhr.pubs.asha.org/pdfaccess.ashx?url=/data/journals/jslhr/929262/ by a Universite Laval User on 06/07/2017
Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

645

to differing, but competent, interpretations of the rules.
It follows, therefore, that the rule-based system designed
for connected speech does not provide the guidance
needed by raters if they are to make reliable judgments
in the complex interactive environment of conversation.
This conclusion is further supported by the fact that 26%
(92) of the 251 disagreements occurred in conversational
contexts not addressed by the rules.
This result is not surprising if it is remembered that
the rules for the CIU were designed to be used on data
collected from single-speaker discourse samples (connected speech) and therefore did not need to address
the interactive issues found in conversation. As well, in
their design of the measure, Nicholas and Brookshire
(1993) used discourse elicitation tasks designed to constrain the possible range of “correct,” and therefore informative, responses. As a result, the rule-based system did not need to address the issue of how a reliable
framework for correctness would be established by raters as they scored samples. Raters were expected to already have a stable framework available to them when
they used the system (provided by a picture sequence,
for example). A quite different situation prevails when
raters are asked to apply the CIU analysis to naturally
occurring conversation. Because no predetermined
framework of correctness exists in conversation, raters
in this study had to develop their own. Presumably, they
did so by capitalizing on the information resource provided by the conversational context. However, without
rule guidance on the use of contextual information, differing interpretations of the informativeness of Ed’s talk
occurred. That is, because no predetermined framework
of correctness exists in conversation, raters in the current study were left on their own, without guidance from
the rules, when determining the correctness of a particular word in the sample. As a result, scores differed
for different raters as they brought their varied resources
and experience to the task at hand.
Reconciliation of Doyle et al.’s (1995) report of high
reliability (.88) when the CIU analysis was applied to
conversation samples with the results of this study is
less straightforward, and several explanatory possibilities may be considered. One possibility is that we were
successful, as intended, in investigating the application
of the CIU analysis to everyday language not previously
studied (e.g., naturally occurring conversation). That is,
connected speech differs from elicited conversation
samples, which differ from naturally occurring conversation. Thus,application of the CIU analysis, designed
only for connected speech, varies in reliability accordingly. However, the comparability of interrater reliability levels in Doyle et al. and Nicholas and Brookshire’s
studies (.88 and .90 respectively) and the dissimilarity
of both with ours (.54) raises the question of whether
Doyle et al.’s conversation samples were “more like” the
646

connected speech of a single speaker than the interactive language displayed in the naturally occurring conversation of our subject. If so and the Doyle et al. study
lacked the framework of correctness that influenced our
results, that would account for differences in results
between studies. A second possibility is that Doyle et
al.’s conversation samples did not differ from ours but
that the raters in their study had experiences that fostered the development of similar frameworks for judging correct information units that our raters did not
have. In Doyle et al.’s study, two authors served as raters. It seems fair to assume that they had a priori knowledge of the conversational topic and some expectation
of content for at least half of the conversation samples
(e.g., those obtained in their “topic constrained” condition). Thus, when faced with having to score a unit of
speech as correct or not correct, without guidance from
the rules of the CIU analysis, they were able to draw on
this knowledge to develop similar frameworks for making judgments. Because naturally occurring conversation is dynamically and interactively directed by participants, knowledge of topic and expected content was
not available to our raters. Thus, they independently developed frameworks that were quite dissimilar. A third
possibility for differences in findings relates to the primary design difference in studies. We used a single subject, whereas all previous studies used group designs. Even
though our subject fell within the range of aphasia severity and type noted in Doyle et al.’s study, it is possible that
his participation in conversation was more challenging to
score than that of the “average” person with aphasia.
Clearly, additional research on the application of the
CIU analysis to the naturally occurring conversation of
other individuals with aphasia is needed—not just to
support the generalizability of the results of this study
but also to clarify differences in results across studies.
This research should include a comparison of the reliability and stability of the performance of single individuals on the original CIU analysis protocol with their
performance in naturally occurring conversation. The
study of normal, ordinary speakers would also offer
meaningful information about applicability of the CIU
analysis to naturally occurring conversation.
Future research is also needed to extend the information provided by the current study. This would include any study that fostered the development of a rule
system designed specifically to address the complexities
of determining concepts of informativeness and correctness in the complex interactive environment of naturally occurring conversation. Findings of this study suggest that a better understanding of how people produce
meaning in a conversation (particularly when one or
more of the speakers is a disordered speaker) is central
to success of any such system. With such an understanding, guidelines can be designed that provide raters with

Journal of Speech, Language, and Hearing Research • Vol. 42 • 636–648 • June 1999

Downloaded From: http://jslhr.pubs.asha.org/pdfaccess.ashx?url=/data/journals/jslhr/929262/ by a Universite Laval User on 06/07/2017
Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

an interpretive framework upon which they can base
their judgments of correctness/informativeness. Even
without this understanding, however, we can expect
that a successful rule system would include rules for
assessing informativeness based on the impact that the
subjects’ utterances have on their conversational partners. An utterance, for example, might be considered
informative if it does not elicit any repair requests from
the conversational partner; if it does elicit a repair request, it might be considered informative if that repair
request is specific rather than general (i.e. “When did
you do that?” as opposed to “What?”). The rules would
need to include guidelines for incorporating the information provided by other speakers; for making decisions
regarding idiomatic, social, and automatic phrases; and
for handling speech that is used by participants to organize their conversational interaction. A successful system would need to include rules addressing question and
answer sequences, interruptions, and other conversational elements.
Clinical application of the CIU analysis is certainly
warranted for assessment of connected speech as described by Nicholas and Brookshire (1993) or the conversation conditions described by Doyle et al. (1995).
However, this study suggests that capturing everyday
language ability as exemplified by the naturally occurring conversation of a person with aphasia with the CIU
rule-based system as currently designed remains a challenge to the field.

Acknowledgments
This study was conducted as a master’s thesis by the
second author that was directed by the first author. We would
like to express our appreciation to Bob Brookshire, Bob
Marshall, and Associate Editor Shari Baum for their objectivity in reviewing this work and for their meaningful comments.

References
Atkinson, J. M., & Heritage, J. (1994). Structures of social
action: Studies in conversation analysis. Cambridge, U.K.:
Cambridge University Press.
Blomert, L., Koster, C., Van Mier, H. & Kean, M-L.
(1987). Verbal communication abilities of aphasic patients:
The everyday language test. Aphasiology, 1, 463–474.
Brookshire, R. H., & Nicholas, L. E. (1995). Performance
deviations in the connected speech of adults with no brain
damage and adults with aphasia. American Journal of
Speech-Language Pathology, 4(4), 118–123.
Brookshire, R. H., & Nicholas, L. E. (1994). Speech
sample size and the test-retest stability of connected
speech measures for adults with aphasia. Journal of
Speech and Hearing Research, 37, 399–407.
Crockford, C., & Lesser, R. (1994). Assessing functional
communication in aphasia: Clinical utility and time

demands of three methods. European Journal of Disorders
of Communication, 29, 165–182.
Doyle, P. J., Goda, A. J., & Spencer, K. A. (1995). The
communicative informativeness and efficiency of connected discourse by adults with aphasia under structured
and conversational sampling conditions. American
Journal of Speech-Language Pathology, 4(4), 130–134.
Doyle, P. J., Tsironas, D., Goda, A. J., & Kalinyak, M.
(1996). The relationship between objective measures and
listeners’ judgments of the communicative informativeness of the connected discourse of adults with aphasia.
American Journal of Speech-Language Pathology, 5(3),
53–60.
Goodwin, C. (1995). Co-constructing meaning in conversations with an aphasic man. Research on Language and
Social Interaction, 28, 233–260.
Goodwin, C. (1987). Forgetfulness as an interactive
resource. Social Psychology Quarterly, 2, 115–131.
Goodwin, M. H., & Goodwin, C. (1986). Gesture and coparticipation in the activity of searching for a word.
Semiotica, 62, 51–72.
Goodglass, H., & Kaplan, E. (1972). Boston Diagnostic
Aphasia Examination. Philadelphia: Lea and Febiger.
Helmsley, G., & Code, C. (1996). Interactions between
recovery in aphasia, emotional and psychosocial factors
in subjects with aphasia, their significant others and
speech pathologists. Disability and Rehabilitation, 18,
567–584.
Holland, A. (1980). Communicative Abilities in Daily
Living. Baltimore: University Park Press.
Kertesz, A. (1982). The Western Aphasia Battery. New York:
Grune & Stratton.
Levinson, S. C. (1983). Pragmatics. Cambridge, U.K.:
Cambridge University Press.
Lomas, J., Pickard, L., Bester, S., Elbard, H., Finlayson,
A., & Zoghaib, C. (1989). The Communicative Effectiveness Index: Development and psychometric evaluation of a
functional communication measure for adult aphasia.
Journal of Speech and Hearing Disorders, 54, 113–124.
Nicholas, L. E., & Brookshire, R. H. (1995). Presence,
completeness, and accuracy of main concepts in the
connected speech of non-brain-damaged adults and adults
with aphasia. Journal of Speech and Hearing Research,
38, 145–156.
Nicholas, L. E., & Brookshire, R. H. (1993). A system for
quantifying the informativeness and efficiency of the
connected speech of adults with aphasia. Journal of
Speech and Hearing Research, 36, 338–350.
Oelschlaeger, M. L. (1999). Participation of a conversation
partner in the word searches of a person with aphasia.
American Journal of Speech-Language Pathology, 8, 62–71.
Oelschlaeger, M. L., & Damico, J. S. (1998a). Joint
productions as a conversational strategy in aphasia.
Clinical Linguistics and Phonetics, 12, 459–480.
Oelschlaeger, M. L., & Damico, J. S. (1998b). Spontaneous verbal repetition: A social strategy in aphasic conversation. Aphasiology, 12, 971–988.
Orange, J. B., Lubinski, R. B., & Higginbotham, D. J.
(1996). Conversational repair by individuals with

Oelschlaeger & Thorne: Application of the Correct Information Unit

Downloaded From: http://jslhr.pubs.asha.org/pdfaccess.ashx?url=/data/journals/jslhr/929262/ by a Universite Laval User on 06/07/2017
Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

647

dementia of the Alzheimer’s type. Journal of Speech and
Hearing Research, 39, 881–895.

and assessment in behavioral sciences (pp. 261–314).
Reading, MA: Addison-Wesley.

Pedhazur, E. J., & Schmelkin, L. (1991). Measurement,
design, and analysis: An integrated approach. Hilldale,
NJ: Lawrence Erlbaum Assoc.

Simmons-Mackie, N. N., & Damico, J. S. (1996). The
contribution of discourse markers to communicative
competence in aphasia. American Journal of SpeechLanguage Pathology, 5, 37–43.

Porch, B. E. (1981). The Porch Index of Communicative
Ability. Palo Alto, CA: Consulting Psychologists Press.
Sacks, H. (1992). Lectures on conversation. Cambridge,
U.K.: Basil Blackwell.

Received January 1, 1998

Sacks, H., Schegloff, E. A., & Jefferson, G. (1974). A
simplest systematics for the organization of turn-taking in
conversation. Language, 50, 361–382.

Contact author: Mary L. Oelschlaeger, PhD, Department of
Speech Pathology and Audiology, Northern Arizona
University, PO Box 15045, Flagstaff, AZ 86011. Email:
mary.oelschlaeger@nua.edu

Seibel, D. W. (1968). Measurement of aptitude and achievement. In D. K. Whitla (Ed.), Handbook of measurement

Accepted August 25, 1998

Appendix. Rater packet.
Instructions:

6. The symbol “*” is used to indicate unintelligible utterances.

This study is an examination of a system for measuring the
communicative informativeness and efficiency of connected
speech of a person with aphasia. Your job will be to apply a
set of rules (included in your packet) to the connected speech of
one participant in two conversations. The rules will generate
two kinds of information: a word count and a count of Correct
Information Units (CIUs). You will count both words and CIUs
from a transcript of the two conversations. Please follow these
steps when recording your word and CIU counts.

7. Please read the complete set of rules prior to beginning
your scoring. There are rules for both words and CIUs.

1. Score only what is said by our subject with aphasia,
Ed (E).
2. Record your count of words in the blank marked “W” at
the end of each of Ed’s turns.
3. Record your count of CIUs in the blank marked “CIU” at
the end of each of Ed’s turns.
4. Do not score any turn which appears in italics. These are
included in order to supply context only.
5. Do not score any words appearing in parentheses. These
are either best guesses at unintelligible utterances or
editorial comments. Again these are provided in order to
provide context.

648

8. I will be keeping track of how long this task takes. You
have as long as you need in order to complete this task. If
you are not finished by the end of the time we have
arranged for today, we will schedule a time when scoring
can be completed.
9. If you need anything from me while you are scoring,
please feel free to ask.
Thank you so much for your help with this study, and good
luck.

Rules for Word and CIU Count:
Raters were also given a photocopy of the rules from the
original article on the %CIU: Nicholas, L. E., & Brookshire, R.
H. (1993). A system for quantifying the informativeness and
efficiency of the connected speech of adults with aphasia.
Journal of Speech and Hearing Research, 36, 338–350. The
rules may be found in Appendix B of their article on pages
348 to 350.

Journal of Speech, Language, and Hearing Research • Vol. 42 • 636–648 • June 1999

Downloaded From: http://jslhr.pubs.asha.org/pdfaccess.ashx?url=/data/journals/jslhr/929262/ by a Universite Laval User on 06/07/2017
Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx